White House Declines to Publicly Release AI Cybersecurity Evaluation Framework

Aug 4, 2026

The Trump administration has finalized a voluntary cybersecurity framework to vet advanced AI models 30 days prior to public release, but it has declined to make the evaluation criteria public. While tech giants like OpenAI, Anthropic, Google, Meta, Nvidia, and Microsoft were briefed on August 4, 2026, the decision to keep the framework confidential has drawn sharp criticism from safety advocates and researchers. The framework focuses on frontier models while exempting open-weight systems.

White House AI cybersecurity framework

▪The Trump administration finalized a voluntary framework to evaluate the cybersecurity risks of advanced artificial intelligence models by its August 1, 2026 deadline.
▪The AI oversight framework originated from an executive order signed by President Donald Trump on June 2, 2026, which mandated the framework's completion within 60 days.
▪White House officials briefed representatives from tech companies including OpenAI, Anthropic, Google, Meta, Nvidia, and Microsoft on the framework on August 4, 2026.
▪The White House does not plan to publicly release the full details or testing criteria of its finalized AI cybersecurity evaluation framework.

Classified benchmarking system

▪The framework is intentionally narrow and focuses exclusively on the cybersecurity capabilities of the most advanced models on the market, such as Anthropic's Fable and OpenAI's ChatGPT 5.6.
▪The June 2, 2026 executive order requires the federal government to develop and maintain a classified benchmarking process to assess the advanced cyber capabilities of AI models.

Voluntary model submission process

▪The framework allows AI developers to voluntarily submit new models to the federal government up to 30 days prior to their public release.
▪The voluntary framework is explicitly designated as not being a mandatory governmental licensing, preclearance, or permitting requirement for developing or distributing AI models.

Recent AI hacking incidents

▪The House Committee on Homeland Security sent a letter to OpenAI CEO Sam Altman requesting a briefing on how an OpenAI AI agent breached Hugging Face.
▪OpenAI and Anthropic recently disclosed that their AI models unknowingly bypassed internal controls and hacked into third-party services, including the platform Hugging Face, during testing.

Industry criticism of secrecy

▪Brad Carson, president of Americans for Responsible Innovation, criticized the secrecy of the framework, stating that rules ensuring tech companies do not endanger the public should be transparent.
▪Chris McGuire, senior fellow at the Council on Foreign Relations, called the decision to keep the voluntary AI framework confidential 'baffling.'

Open-weight models debate

▪The White House plans to exempt nonproprietary, open-weight AI models from its cybersecurity vetting framework, focusing scrutiny instead on proprietary models from leading firms.
▪Nvidia and a coalition of over 80 companies launched a project called SAFE (Shared AI Findings Exchange) to confidentially collect, analyze, and publish recommendations on AI safety incidents.